BGP scaling revisited
Brian Carpenter
Department of Computer Science
The University of Auckland
November 2011
Work partially supported by
What's the BGP question again?
●
Are there any interesting long-term
relationships between the size of the globally
addressable Internet and the size of the BGP4
system?
What data do we have?
●
We have BGP4 data back to 1994 and active
AS data back to 1997 (thanks potaroo.net)
●
We have domain count data back to 1994
(thanks ISC.org)
–
the domain count is a reasonable lower bound on
the number of directly accessible IPv4 interfaces
with global addresses
–
if you don’t agree with this assertion please talk to
ISC about it ;-)
–
(for earlier years we have actual host counts)
Domain count history
850000000
800000000
750000000
700000000
650000000
600000000
550000000
500000000
450000000
400000000
350000000
300000000
250000000
200000000
150000000
100000000
50000000
0
6777777777788888888889999999999000000000011
9012345678901234567890123456789012345678901
Data from http://www.isc.org/
Domain count history knitted onto host
count history (log scale)
1000000000
100000000
10000000
1000000
100000
10000
1000
100
10
1
6777777777788888888889999999999000000000011
9012345678901234567890123456789012345678901
Data from http://www.isc.org/ and other sources
slope = 7.78
1997-2008
19 routes/AS
9.5 routes/AS
slope = 8.98
1997-2011
9.6 routes/AS
slope = 9.84
1994-2008
slope = 11.9
1997-2011
slope = 0.75
1994-2008
slope = 0.70
1997-2011
What do we learn?
●
BGP4 as a whole is scaling something like the
square root of the total size of the addressable
Internet.
●
This is a Good Thing.
–
●
It could have been so much worse.
But it only holds true as long as operational
policy and habits don’t change.
–
Complacency is not in order.